AI-Driven Automated Scientific News Generation and Dissemination System Using Large Language Models and Telegram Integration

THEK Research Institute — White Paper Series | Version 1.0 | 2026 Author: Marcelo Moncayo Theurer, M.Sc. (TUM) — CSO, THEK Research Institute | Full Professor, University of Guayaquil, Ecuador

This white paper presents the design, implementation, and results of an AI-driven autonomous scientific news generation and dissemination system developed at THEK Research Institute (Guayaquil, Ecuador). The framework integrates large language models (LLMs) — specifically OpenAI GPT-4.1-mini and Google Gemini 2.5-flash — with real-time RSS feed processing, Python automation, programmatic image generation (Pillow), and the Telegram Bot API into a unified, fully autonomous scientific communication pipeline.

The system autonomously retrieves scientific topics from authoritative sources (ScienceDaily, NASA, Phys.org, Archaeology Magazine, LiveScience), synthesizes 300-word educational scientific summaries, produces branded multimedia imagery, and publishes formatted posts to a Telegram channel — all without human editorial intervention. Scientific domains covered include astronomy, geophysics, archaeology, mathematics, advanced instrumentation, and engineering.

Two operational implementation variants are documented: 

• Variant A (GPT-4.1-mini): Prompt-driven generation of 10 science news articles per cycle from configurable topic domains. Uses python-telegram-bot SDK for publication.
• Variant B (Gemini 2.5-flash + RSS): Live ingestion from 5 authoritative RSS feeds with LLM-based synthesis grounded in real source content. Includes canonical source URLs in every published post.

Both variants were tested on standard hardware and confirmed successful autonomous operation across multiple execution cycles with cycle times of 90–140 seconds per batch.

Key contributions:

Fully functional, reproducible autonomous scientific communication prototype

Two LLM backend implementations (OpenAI / Google Gemini) with detailed code documentation

Modular pipeline architecture extensible to RAG, multilingual output, and multi-platform broadcasting

Strategic roadmap toward AI-assisted seismic alert systems relevant to Ecuador and the Pacific Ring of Fire

 Open-science posture with CC BY 4.0 license for unrestricted reuse.

Future directions include retrieval-augmented generation (RAG), credibility verification, Spanish-language dissemination, multi-platform broadcasting (X, Instagram, LinkedIn, WhatsApp), and real-time geophysical event reporting integrated with IGEPN and USGS data streams.

About the Author: Marcelo Moncayo Theurer, M.Sc. (TUM), is the CSO of THEK Research Institute and Full Professor at the University of Guayaquil (Faculty of Physics and Mathematics). His research spans seismology, geophysics, tectonics, earthquake engineering, catastrophe modeling, and computational scientific communication. He has directed the design of Ecuador's largest tunnel and longest bridge, supervised over 60 undergraduate theses, and is a regular science columnist for El Universo, Ecuador's leading national newspaper.